Classification of a Population of Objects by Convolutional Dictionary Learning with Class Proportion Data

Unmet Need
White blood cells (WBCs) -- which consist of granulocytes, lymphocytes, and monocytes -- are essential to the body’s immune system. Deficiency in any type of WBC can lead to infection or inflammation, and knowledge of a patient's WBC count is essential for the diagnosis and treatment of a wide range of conditions, including leucopenia, insufficient WBC, and proliferative disorders (excessive WBC). The Complete Blood Count (CBC) is the most commonly requested blood test by doctors, as it provides the concentrations of various types of cells in a blood sample. 
Given high resolution images of stained WBCs, they can be differentiated by their differences in size and nuclear shape, and traditional methods such as Support Vector Machines (SVMs) or Dictionary Learning based methods can be successfully used to classify WBCs. However, traditional image-based classification methods do not perform well when dealing with low resolution images of unstained WBCs. There is therefore a need for a novel method that is able to accurately count and classify unstained WBCs in the low resolution setting, as such a method is key to developing new technology to perform a low-cost CBC.
 
Technology Overview
Johns Hopkins researchers have developed a method for classifying WBCs from images of lysed blood from both normal and abnormal blood donors (lysed blood contains WBCs, in addition to debris from lysed RBCs). Using a generative model that was trained using both purified WBC images and CBC results from more than 300,000 patients at the Johns Hopkins hospitals, the method can differentiate WBCs into granulocytes, lymphocytes, and monocytes, as well as count the total number of WBCs in the image. This technology enables counting and differentiation of WBCs, even in low resolution images of unstained WBCs.  This method has the potential to enable new, low-cost and rapid CBC technology.
 
Stage of Development
The invention has been tested with 36 lysed blood samples from both normal and abnormal donors. The strongest correlation between the predicted and ground truth cell concentrations appeared for granulocytes and lymphocytes. The monocyte correlation was not as strong, however the absolute errors between the predicted and ground truth monocyte concentrations were small.

Category(s):